Single Shot MultiBox Detector utilizes a single network to identify as well as classify the objects in an image.
MultiBox
In Single Shot MultiBox Detector, MultiBox is the name of the technique for bounding box regression developed by Szegedy et al.
MultiBox proposes coordinate for the bounding box.
MultiBox gives two important elements:
Location Loss(l): This corresponds to the difference between the predicted bounding box and ground truth bounding box.
Confidence Loss(c): It is a measure of how sure the network is, of object belonging to a particular class.
Multibox box Loss(m), which is how correct our overall prediction is given by: m = c + αl
Where α is used to balance the contribution of location loss.
Jaccard Index = Area of Overlap of Bounding box/Area of Union Of Bounding Box. MultiBox endeavors to relapse nearer to the ground truth, but detection begins from prior.
Code in this repository contains Implementation Of SDD in Pytorch
- Activate the virtual environment.
- Download and load the weight(in object_detection.py)<I'll create a link for weight>
net.load_state_dict(torch.load('/home/dragonbreath/Zenith/Python/Projects/Object Detection/ssd300_mAP_77.43_v2.pth',map_location = lambda storage, loc: storage )) - Load the video File(in object_detection.py):
reader = imageio.get_reader('/home/dragonbreath/a.mp4') - Run object_detection.py via
python object_detection.py - It will process all the frames in the video and generate the video output of detected objects.
Code supports both GPU(if your have cuda enabled) and CPU
